AI weekly roundup · 2026-W40

AMD announced that it has signed a definitive agreement to acquire World Labs, the spatial intelligence company founded by Fei-Fei Li, in an all-stock transaction valued at about $8.2 billion. The deal has not closed. It is expected to complete by the end of 2026, subject to regulatory approval and other customary closing conditions; after closing, Fei-Fei Li will become AMD's Executive Vice President and Chief Scientist, reporting to Lisa Su.

OpenAI released GPT-6.1 Sol, saying it delivers near-GPT-6-Astra intelligence for coding, computer use and professional work at one-fifth of Astra's standard API input and output token prices. GitHub Copilot, Amazon Bedrock and Codex CLI added it the same day.

Google introduced Gemini 4 Argon, a frontier model for complex real-world coding, enterprise knowledge work, and cyber defense with an output limit raised to one million tokens. It is now rolling out to trusted cyber defenders through the Fairwind Program, with an introductory price of $2 per million input tokens and $10 per million output tokens.

Aleph Alpha released Kolibri-1 on October 3. The Apache 2.0 MoE model has about 78B total parameters and about 3.46B active per token, supports tool calling, and offers a maximum context of 1,048,576 tokens with a lower recommended operating length.

Anthropic released Claude Sonnet 5.5, the second model in the Claude 5.5 family, at the same list price as Sonnet 5: $2 per million input tokens and $10 per million output tokens. Anthropic says its output is more than 30% faster than Sonnet 5 and that, in its own testing, the cost per task is up to 30% lower (mainly because it needs fewer tokens to do the same work). The model is available on AWS, Google Cloud and Microsoft Azure.

Anthropic introduced Claude Code Mods: TypeScript functions that change prompts, tool calls, and the interface. Mods ship in plugins for the CLI and desktop app, run with user permissions, and are not sandboxed.

GitHub added a computer-use public preview to Copilot CLI and the Copilot app on macOS and Windows. It can read app content, click, type, and navigate workflows across apps, with approval before app control and organization settings that can disable it.

OpenAI published a guide for choosing among GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna. It connects task difficulty with model choice and recommends evaluating task success, latency and cost in production workflows.

Anthropic released Claude Code 2.1.289 with fixes to Bash deny and ask rules, IDE symlink Read restrictions, and managed MCP sign-in tool descriptions. The release also adds agent.spawn for teammates.

GitHub added dynamic workflows to Copilot CLI, the app, and SDK. Code combines automated steps and agents with sequential or parallel execution and checkpoint pauses; all Copilot plans are supported, while the CLI needs experimental features enabled.

AWS published a SageMaker AI workflow for training a Qwen3.6-27B search agent with multi-turn reinforcement learning. It uses retrieval tools and trajectory rewards, and reports gains on three of four evaluated held-out sets.

AWS published a tutorial connecting Claude Desktop configured for Amazon Bedrock to AgentCore Gateway’s MCP web search. The example combines IAM Identity Center, Cognito and OAuth authentication with AWS-managed search.